Incomplete observation process

[afmagee42]

The current implementation assumes perfect observation of all cases. Generalizing beyond this would be nice.

First-pass is probably to assume cases are observed independently with probability $\rho$, yielding
$\text{Pr}(j \mid R, k, \rho) = \sum\nolimits_{\ell = j}^{\infty} \text{Pr}_{\text{Binomial}}(j \mid \ell, \rho) \text{Pr}_{\text{NBBP}}(\ell \mid R, k)$
We may also want to set upper limits on $\ell$, which contain some amount of information, via "I really don't think we could have missed more than $\ell - j$ cases."

We will have to give some thought as to how (and whether) to combine incomplete observation with censoring.